Sie sind auf Seite 1von 4


discussions, stats, and author profiles for this publication at:

Nonlinear Programming

Article in Journal of the Operational Research Society · January 1995

DOI: 10.1057/palgrave.jors.2600425


4,264 2,077

1 author:

Dimitri P. Bertsekas
Massachusetts Institute of Technology


Some of the authors of this publication are also working on these related projects:

1) Approximate and abstract dynamic programming. 2) Proximal algorithms for large-scale linear
systems of equations View project

All content following this page was uploaded by Dimitri P. Bertsekas on 21 December 2016.

The user has requested enhancement of the downloaded file.

Journal of the Operational Research Society (1997) 48, 332±334 #1997 Operational Research Society Ltd. All rights reserved. 0160-5682/97 $12.00

Book Selection
Edited by JOHN M. WILSON

CHARNES A, COOPER W, LEWIN AY and SEIFORD LM (eds). Data Envelopment Analysis Theory, Methodology
and Applications 332
KORSHUNOV AD (ed). Discrete Analysis and Operations Research 333
BIETHAHN J and NISSEN V (eds). Evolutionary Algorithms in Management Applications 333
AVRIEL M and GOLANY B (eds). Mathematical Programming for Industrial Engineers 334
BERTSEKAS DP. Nonlinear Programming 334

Data Envelopment Analysis Theory, Methodology and researchers to gain an entree into the subject, and to use a
Applications common terminology in describing their work. Indeed in a
noteworthy development the publishers will give permis-
CHARNES A, COOPER W, LEWIN AY and SEIFORD LM (eds). sion to authors to incorporate material from these early
Kluwer Academic Publishers, London, 1995, xii ‡ 513 pp. chapters into their books. The case studies in Part II cover a
£105.00 ISBN 0 7923 9479 8 wide range of topics, and are written by many of the most
prominent researchers in DEA. They will undoubtedly
The pursuit of organizational ef®ciency has become some- inspire both researchers and practitioners. In Part III
thing of an obsession for the modern manager. However, almost all readers will offer grateful thanks for Larry
managers interested in identifying ef®cient practice have Seiford's remarkably comprehensive bibliography.
few tools available, other than the juvenilia hawked by The editors hope that the text will serve as an introduc-
accountants. A notable exception is the technique known as tion to new users and a reference for the more experienced.
data envelopment analysis (DEA), which was introduced to How does it measure up to its ambitions? It must be said
the world in its modern form in 1978. The deceptively that teaching DEA can give rise to unexpected dif®culties.
simple DEA model has since become a remarkably fertile While the underlying concepts are apparently simple, the
®eld for both research and practice. intricacies of the technique can often prove elusive, parti-
DEA offers an insight into the relative ef®ciency of cularly to those trained in traditional statistical techniques.
comparable `decision-making units' such asÐsayÐ The editors have chosen to present the (traditional) DEA
schools. It offers a conservative estimate of comparative model as the culmination of more general concepts, which
ef®ciency in situations in which multiple inputs and multi- in my judgement is very successful. Although `old hands'
ple outputs are found. The technique exploits limited data may ®nd the ordering of the material super®cially perverse,
to its full extent, and can be used to examine technical I believe the book succeeds in presenting a dif®cult
ef®ciency, allocative ef®ciency and scale ef®ciency. It has technique in a logical and accessible fashion. No serious
proved to be immensely useful in a wide range of settings. DEA researcher should be without the text, and indeed the
On the downside, the technique is sensitive to data accu- project should serve as a model for authors with similar
racy, and some DEA practitioners are guilty of stretching ambitions in other disciplines.
their analysis beyond the capacity of the data available to The book arises from a conference on DEA held in 1989,
them. implying a massive delay in bringing the material to
This long awaited book seeks to summarize and codify publication. While not entirely satisfactory, the hiatus has
the state of current scienti®c knowledge relating to DEA. It resulted in some substantial bene®ts. First, it has given the
comprises three parts: an introduction to the concepts, editors the opportunity to review all the material rigorously,
models and computational aspects of DEA; a set of case in contrast to many conference proceedings. Second, the
studies; and an epilogue and bibliography. Part I ®rst offers book is free of the rather exotic developments that have
an intuitive insight into DEA and then develops a sequence in®ltrated some of the DEA literature in more recent years.
of basic theoretical models, followed by a series of re®ne- Perhaps the one question mark concerns the price the
ments and extensions. The intention is to present a uni®ed publishers have chosen to attach to the book. Even in
treatment of the topic. This part of the book will enable paperback, this is not a book for the impulse buyer.
Book Selection 333

Researchers in DEA will probably have no hesitation in tion Strategies and Evolutionary Programming. The book
®nding the cash, but the price may deter masters students, comprises three sections:
which is a shame becauseÐin spite of its limitationsÐ
 a tutorial on Evolutionary Algorithms by Nissen and
DEA is one of the most distinctive and useful developments
Biethan, which explains how the algorithms work and
to have emerged from the management science community
the differences between them;
in recent years.
 a survey by Nissen of the uses to which evolutionary
University of St. Andrews SMITH PC algorithms have been put; this lists and classi®es some
500 papers up to 1995 and would be an invaluable
starting point for anyone considering using an Evolu-
Discrete Analysis and Operations Research tionary Algorithm;
 19 papers describing various applications of Evolutionary
KORSHUNOV AD (ed). Algorithms classi®ed by domain: industry; trade; ®nancial
services; traf®c management; and (school) timetabling.
Kluwer Academic Publishers, London, 1996, vi ‡ 343 pp.
£119.00 ISBN 0 7923 3866 9 Evolutionary Algorithms are heuristic search and opti-
mization techniques which are motivated by analogies with
The book has many interesting papers in it. Many of these the natural processes of evolution. They bear some simila-
topics will be found useful by operational research analysts. rities to Simulated Annealing in that the search process is
On the other hand many of the topics are better suited for a stochastic and can go backwards (i.e. the objective function
computer scientist. The book is divided between what is can worsen) so as to move away from the current area of
considered operational research and what might be consid- search. The big difference is that whereas Simulated
ered computer science subject matter with the greater Annealing works with a single incumbent solution, Evolu-
emphasis on the latter. For example, the paper entitled An tionary Algorithms use a population of incumbent solutions
Approximation Algorithm for the Traveling Salesman (a generation) which are bred together to form the next
Problem and its Probabilistic Analysis is a topic that one generation of incumbent solutions.
one might expect in a book with Operations Research in its One of the keys to success with Simulated Annealing is
title. On the other hand the topic The Number of Distinct de®ning the topology of solutions, namely which solutions
Subwords of Fixed Length in the Morse-Hedlund Sequence are in the neighbourhood of the incumbent solution. Simi-
would be a subject near and dear to the heart of a computer larly with Evolutionary Algorithms one must characterise
scientist interested in discrete mathematics which should be potential solutions and specify the permitted transitions
no surprise as it is also in the title of the book. between them. But here the evolutionary paradigm comes
Being a student of both operational research and discrete to the fore, especially with Genetic Algorithms. A solution
mathematics, I found the book interesting and informative. is characterised by its genotype, which is a coding of the
This of course being a matter of individual preferences. I solution as a bitstream, and its ®tness, namely objective
am not certain that everyone shares my interests. Hence I function value. Two solutions are selected for breeding
recommend the book with a note of caution. If you enjoy depending on their ®tness and their genotypes are mixed
both operations research and discrete mathematics, you together to form two offspring, each of whose genotypes
should enjoy this book. If you are searching for a research may then mutate with some probability. The resulting
topic, this book would provide a number of possibilities. offspring of all the parents form the new generation.
On the other hand if you do not care to wade through a The key problems in developing Evolutionary Algo-
great deal of mathematics for what might be a gem or two, rithms are therefore working out how to encode solutions
this is not the book for you. as genotypes and what exchanges of genetic material
should take place at breeding. It is because of this that
University of Maryland Leake C the second and third parts of this book are so useful: the
latter presents a wide a variety of papers to stimulate one's
thoughts while the former provides a starting point for
Evolutionary Algorithms in Management Applications further research.
As an instance of what can be achieved, the paper by
BIETHAHN J and NISSEN V (eds). Falkenauer describes using a Genetic Algorithm to tackle
Springer, Berlin, 1995, xv ‡ 378 pp. DM 148.00 the Bin Packing Problem: given a number of items of
ISBN 3 540 60382 4 different sizes, what is the smallest number of bins of a
particular size into which they can be packed? The standard
This is an excellent book. It presents the ®eld of Evolu- encoding techniques for Genetic Algorithms are not suita-
tionary Algorithms, which embraces Genetic Algorithms as ble and Falkenauer develops an elegant encoding which
well as the less well-known Genetic Programming, Evolu- avoids the problem of redundancy (this occurs where a
334 Journal of the Operational Research Society Vol. 47, No. 1

single logical solution has many distinct genotypes). He mathematically able students (the standard of numeracy
then uses a local optimization algorithm in the course of expected is high) who are looking at optimisation models in
breeding and ends up with a hybrid algorithm which O.R. or industrial engineering, this is a book to look at
combines the best features of Genetic Algorithms with seriously.
speci®c bin-packing algorithms of Martello and Toth.
University of Exeter Smith DK
This paper is valuable not only for the results which it
achieves but also for showing the limitations of Genetic
Nonlinear Programming
Algorithms and how the ideas of Genetic Algorithms can
be combined with other techniques where appropriate. BERTSEKAS DP
Altogether this book is well worth buying if you have Athena Scienti®c, Belmont, Mass., 1995, x ‡ 646 pp.
any interest in using Evolutionary Algorithms. $69.00 ISBN 1 886529 14 0
Robert Simons Ltd Simons R
In the June 1996 issue of this Journal1, I reviewed `Dynamic
Mathematical Programming for Industrial Engineers
Programming and Optimal Control' by the same author,
AVRIEL M and GOLANY B (eds). from the same publisher. It was very tempting to repeat the
same review for this book with minor alterations. This, like
Marcel Dekker, New York, 1996, 648 pp. $175.00
the earlier text is a work ®lled with a detailed mathematical
ISBN 0 8247 9620 9
presentation of the subject. This, too, is presented in such a
user-unfriendly way as to diminish the value of the contents
In this textbook for industrial engineering students, math- enormously.
ematical programming is divided into eight parts. Predic- Bertsekas presents optimization of nonlinear functions as
tably, linear programming and integer programming lead an essentially mathematical problem. Given an objective
the way, (over 100 pages each), followed by network function, one wishes to maximise it by a suitable search
problems, dynamic and nonlinear programming. The tail method using as much information as possible about the
end (a quarter of the book) is devoted to multiobjective and behaviour of the functionÐand that information comes,
stochastic programming and a short chapter on heuristics. almost certainly as far as this book is concerned, from
The strength of the work is that each of these chapters mathematical analysis. Some concessions are made to the
has been written by different authors, so that one is spared problem of computing, but these are scant, and for practical
the patchy nature so often encountered when an expert in purposes the user would have to turn to other texts.
one ®eld attempts to pad out a book with some obligatory The author turns from an opening chapter on uncon-
chapter describing other ®elds. Looking at the result, it strained problems to constrained ones, which allows him to
seems that the editors had given the separate authors very include interior point methods for linear programming. This
strong guidelines about the structure of each chapter, so is followed by extensive treatment of Lagrange multiplier
that there is a pleasantly uniform treatment of each topic. theory and applications, and the ®nal two chapters of the
As a book intended for classroom use, each chapter has a book cover duality theory, again with a theoretical treat-
good mixture of theory and practice, with some examples ment of the applications of this. Four appendices provide
and exercises to develop the ideas presented in the text. some essential mathematical background for the reader.
With an eye to the readership of the title, many examples The strength of this book is in its comprehensiveness.
are relevant to industrial engineeringÐwith a few notable Where the author has chosen to include a topic, then it is
exceptions! well-covered. HoweverÐand that is a word that this
The fact that this is a collection, however strong the reviewer wanted to use much more frequently in this
editorial hand, also leads to its weakness. The authors pieceÐthere are omissions. Very little reference is given
overlap in their treatment of some optimization problems, to the signi®cant contributions to the subject that have been
but there is no cross-referencing between chapters. Some made by U.K. based researchers, and to the needs of users
topics are left out completely. The chapters make different of algorithms. The treatment of line searches is absurdly
assumptions about the depth of mathematical treatment brief; methods based on complementarity do not appear.
needed in each topic, and there are variable numbers of I will use my copy to help in teaching a course on
references to cases and further books and papers. The book nonlinear optimization, but it will not be the ®rst book that I
has a generally good subject index. turn to for such help.
At its published price, this is a book with limited appeal
to individuals (and even to academic libraries). For student University of Exeter Smith DK
use, the price would have to be much lower to compete
with any of the well-established general operational Reference
research textbooks supplemented by directed reading for 1 Smith DK (1996). Review of Dynamic Programming and
a more thorough treatment, but if you have a group of Optimal Control by Bertsekas DP J Opl Res Soc 47: 833±834.

View publication stats